In one embodiment, techniques for optimizing voice quality for a media stream provided to participants in a conference are provided. A first path is provided that provides an unmixed path of media from a first user to participants in the conference. A second path includes a mixed path of media from multiple users. The mixed path may include media from a first user and a second user that is mixed. The mixed path may include a transcoding from a first compression to a second compression, and back to the first compression (or another compression). A floor control module may switch between the first path and the second path to optimize voice quality.